Digital Twin of a Horizontal Three-phase Separator in an Offshore Oil Extraction and Processing Platform using NARX Neural Networks

Abstract

Offshore oil extraction presents several challenging scenarios from exploration to operation. Limitations in physical space impose restrictions on the construction of vessels that have sufficient capacitance to absorb the accentuated load variations to which this process is subject. This factor, allied to the turbulent nature of the flow into the risers, which causes occasional slugs, brings high complexity to treatment and separation vessels level control, upon which the quality of the final products depends. The present work takes advantage of recent advances in process identification techniques to develop a NARX-type neural network to carry out a Digital Twin implementation of a three-phase separator of an offshore unit, located in the Campos basin, Brazil. The Digital Twin allows simulations of different control techniques to be tested in a realistic simulation environment, as it uses heuristics and machine learning techniques capable of inferring even nonlinear relationships between variables, mirroring the physical twin behavior. The main goal is to enable these simulations to be run and achieve enhanced process control. The results obtained in the present work show that it is possible to obtain a Digital Twin using NARX-type neural networks with Mean Absolute Percentage Error marks below 1% in test situations to predict main chamber and oil chamber levels, which can be used to simulate and benchmark advanced level control strategies.